Separating common from distinctive variation.

Area of Science:

• Multivariate data analysis
• Bioinformatics
• Systems biology

Background:

• Joint and individual variation explained (JIVE), distinct and common simultaneous component analysis (DISCO), and O2-PLS are latent variable methods for integrated analysis of multiple datasets.
• These methods decompose data into common variation, dataset-specific variation, and residual noise.

Purpose of the Study:

• To compare JIVE, DISCO, and O2-PLS based on their mathematical properties.
• To analyze how these methods define and capture common and distinctive variation.
• To evaluate their performance on simulated and real biological data.

Main Methods:

• Comparative analysis of JIVE, DISCO, and O2-PLS.
• Application on simulated datasets.
• Application on mRNA and miRNA datasets from Glioblastoma Multiforme (GBM) tumors.

Main Results:

• All methods successfully identify common variation when it is abundant.
• Real-world biological data complexities are handled differently by each method.
• Discrepancies arise in how each method addresses data complexities.

Conclusions:

• JIVE, DISCO, and O2-PLS offer distinct approaches to estimating common and distinctive variation, each with strengths and weaknesses.
• Orthogonality assumptions in these methods may lead to misinterpretation of non-orthogonal biological phenomena.
• The choice of method impacts the interpretation of biological insights from multi-omics data.